UD for NArabizi
Tokenization and Word Segmentation
-
Spaces, commas and periods are generally used as word delimiters. Algerian Arabic being a dialect that contains a lot of code-switching, tokenization rules usually follow the target languages, namely Modern Standard Arabic (MSA), Algerian Arabic and French. Exceptions are often due to the specificities of user-generated content.
-
For French, see the French documentation page. However, contracted prepositions are not considered multiword tokens, which is often the case in French treebanks. Thus, prepositions such as au are not split into à+le, but their particular status are specified in the XPOS and MISC fields.
-
For MSA and Algerian Arabic,see the Arabic documentation page. However, definite articles are considered as tokens, so when they are not delimited by a space,they are split. This often happens with nouns starting with sun letters. For instance, essalam –> e+ssalam. The same rule applies for prepositions when they are governed by nouns, proper nouns and adjectives (fdoute –> f+doute), and it also applies recursively (bilaranib –> bi+l+aranib).
-
Non-final punctuation symbols (such as - and / ) are considered as tokens when they are used for football match scores and fractions (i.e. 3-0 -> 3 - 0, 100/100 -> 100 / 100 )
Morphology
Tags
- All universal POS categories except SYM are used. Arabic Negation particles (both in MSA and Algerian Arabic) such as ma and la, and the restrictive particle ghir are tagged as PART.
- The following verbs are tagged as AUX when they are used for composed tense and aspect constructions:
- the French auxiliairies être et avoir
- The Arabic auxiliary kan
- The Algerian Arabic auxiliary rah
Features
- Verbs have Gender (values: Masc and Fem), Number (values: Sing and Plur) and Person (values : 1,2 and 3) features. Verbs also have a Mood feature (values: Imp and Sub) for imperatives and for one specific subjunctive construction (vive/tahia/viva + nominal)
- Pronouns have a PronType feature (values : Dem, Int and Rel).
- Pronouns have an AdpType feature with the single value Prep.
- The polarity feature applies to particles and adverbs and verbs with the single value Neg.
Syntax
- For Arabic, see the Arabic documentation page. However, the language specific deep relations are not used.
- For French, see the French documentation page. However, the language specific deep relations are not used. Some deep relations are specified in the MISC fields (expl:subj,expl:comp and conj:coord)